Modelling human height and weight: a Bayesian approach towards model comparison

Background/Objectives: Given the availability of large longitudinal data sets on human height and weight, different modelling approaches are at hand to access quantities of interest relating to important diagnostic aims. Subjects/Methods: Statistical modelling frameworks for longitudinal data on human height and weight have to consider the issues of individual heterogeneity and time dependence to provide an accurate statistical characterisation. Further, missing values inevitably occurring within longitudinal data sets have to be addressed adequately to allow for valid inference. The Bayesian framework is illustrated to facilitate stringent comparison of available non-nested model frameworks addressing these issues using simulated and empirical data sets. Results: Comparing random-effects and fixed-effects modelling approaches with the Preece–Baines (PB) model reveals that, for simulated data, the Bayesian approach towards model comparison is effective in discriminating between different model specifications. With regard to analysis of 14 longitudinal data sets, the implicit trade-off between model fit, that is, description of the data, and a parsimonious parameterisation favouring prediction is often best addressed via the PB model. Conclusions: The Bayesian approach is illustrated to allow for effective comparison in case model specifications for longitudinal data are not linked directly via parametric restrictions.